Evaluation of the Quality of Life Associated with Zidovudine Treatment in Asymptomatic Human Immunodeficiency Virus Infection

The design and results of Protocol 019 have been reported elsewhere². A double-blind, randomized, placebo-controlled clinical trial, the study was designed to evaluate the efficacy and safety of two doses of zidovudine (1500 mg daily and 500 mg daily) in asymptomatic HIV-infected patients and did not focus on the quality of life. Patients were randomly assigned to treatment groups in two strata according to the CD4+ cell count (<200 and 200 to 499 cells per cubic millimeter). The current analysis considers the same 1338 patients and end points analyzed in the original report of the study findings and includes data on adverse events from May 10, 1989, through August 10, 1989, which were not in the original report.

Adverse events, whether or not they were related to treatment, were classified as laboratory findings or symptoms according to a system previously developed⁹. Laboratory findings were events that could be verified only by a laboratory analysis. Hepatic abnormalities (for example, elevated liver-function values), hematologic abnormalities (for example, neutropenia), and metabolic abnormalities (for example, hypoglycemia) were grouped as laboratory findings.

Symptoms were adverse events with a definite subjective component experienced by the patients, which therefore had a direct effect on the quality of life. (Severe subjective manifestations of objective laboratory findings, such as fatigue associated with anemia, were documented as symptoms.) Symptoms were classified primarily on the basis of organ systems, as follows: gastrointestinal, psychiatric, rheumatologic, dermatologic, constitutional, neurologic, cardiac, genitourinary, otorhinolaryngologic, gynecologic, respiratory, and ophthalmologic symptoms; malaise and fatigue; headache; fever; and other or uncodable events.

To calculate the frequency of adverse events, we counted only the first occurrence of an event in a defined category for a given patient. Thus, the count within each category represents the number of patients (not the number of events). For example, a patient's report of cough and shortness of breath was counted as one respiratory event. Patients who reported adverse events in more than one category were counted in each. Only severe (grade 3) or life-threatening (grade 4) events were included. Odds ratios and associated 95 percent confidence intervals were calculated by standard methods¹¹. Differences between frequencies of events were calculated without correction for multiple comparisons.

The data came from an events file that contained all reported adverse events recorded in the study data base. Events were graded according to the standard system for grading toxicity outlined in the protocol. For example, in the gastrointestinal category, vomiting requiring intravenous therapy was given a grade of 3, and intractable vomiting a grade of 4. Depending on the type or degree of adverse event, the drug (whether zidovudine or placebo) was reduced in dose, temporarily withheld, or permanently stopped.

The quality-of-life-adjusted treatment comparison was based on a modification of a method originally developed to evaluate adjuvant therapies for breast cancer^12,13. This method compares treatments by calculating the period during which patients have no severe symptoms and no progression of disease. To assess individual preferences for zidovudine as compared with placebo, a threshold-utility analysis^9,12-14 was used, which incorporated the weights that reflected the relative reduction in the quality of life associated with adverse events on the one hand and disease progression on the other.

Three periods during the course of follow-up in individual patients were identified: the time without symptoms of disease or toxicity (TWiST), calculated as the number of months preceding the development of a symptom with a grade of 3 or higher or the progression of HIV disease, whichever occurred first; the period after the first occurrence of an adverse event with a grade of 3 or higher (AE); and the period after the progression of HIV disease (Prog). A quality-of-life-adjusted survival model was constructed with the use of utility coefficients u_AE and u_Prog to reflect the value of the state of health (time after an adverse event and time after disease progression, respectively) on a utility scale with reference points of 0 and 1: TWiST was assigned a weight of 1, and death a weight of 0^9,12-14. The quality-of-life-adjusted survival relative to TWiST (Q-TWiST) was calculated as follows:

Q-TWiST = TWiST + (u_AE × AE) + (u_Prog × Prog).

Estimates of the average number of months of Q-TWiST for the three treatment groups were determined by the following procedure. For each group separately, Kaplan-Meier curves for event-free survival and progression-free survival were used to partition the overall survival time during the 18 months of follow-up. Event-free survival was the time from enrollment in the study to the development of a grade 3 or higher symptom or the progression of HIV disease, whichever occurred first. Progression-free survival was the time from enrollment to the progression of disease. Kaplan-Meier estimates of the average periods after an adverse event, after the progression of disease, and without either an adverse event or disease progression were calculated separately for each treatment group. These estimates were represented by the areas between the curves.

For any assumed values of u_AE and u_Prog (ranging from 0 to 1), the Q-TWiST for each treatment group was calculated with the equation given above. A threshold-utility analysis was performed to determine the range of u_AE and u_Prog values for which the Q-TWiST was longer with zidovudine treatment and the range of values for which it was longer with placebo. Ninety-five percent confidence intervals and two-sided P values were calculated with a z-test based on observed differences between treatments, with standard errors calculated by the bootstrap method.

To determine whether our results were affected by the initial CD4+ count, we conducted separate Q-TWiST analyses in each stratum and also performed a proportional-hazards regression for Q-TWiST¹⁵.

Figure 1. Figure 1. Estimated Time to the Progression of Disease, According to Treatment Group.

The 18-month progression-free survival was 89 percent (95 percent confidence interval, 85 to 92 percent) for the patients receiving placebo (dotted line), 94 percent (95 percent confidence interval, 90 to 97 percent) for those receiving 500 mg of zidovudine (dashed line), and 92 percent (95 percent confidence interval, 88 to 96 percent) for those receiving 1500 mg of zidovudine (solid line).

As reported previously, the mean duration of follow-up was 61, 55, and 51 weeks for the placebo, 500-mg, and 1500-mg groups, respectively². Disease progression occurred in 38 of the 428 patients in the placebo group, in 17 of the 453 in the 500-mg group (P = 0.01; relative risk, 0.49; 95 percent confidence interval, 0.28 to 0.86), and in 19 of the 457 in the 1500-mg group (P = 0.10; relative risk, 0.65; 95 percent confidence interval, 0.37 to 1.12) (Figure 1). These data correspond to 7.6, 3.6, and 4.3 end points (progression of disease) per 100 person-years of observation for the placebo, 500-mg, and 1500-mg groups, respectively. There were no deaths until after the first end point, although eight subjects died after the development of the acquired immunodeficiency syndrome (AIDS): four in the placebo group, one in the 500-mg group, and three in the 1500-mg group. The death in the 500-mg group was by suicide, and the others were from physical complications of AIDS.

Table 1. Table 1. Severe Adverse Events (Grade 3 or Higher) in Patients Receiving Zidovudine (1500 or 500 mg) or Placebo, during 18 Months of Follow-up.

Table 1 shows the numbers of patients with severe or life-threatening adverse events of various types, according to the treatment group, during 18 months of follow-up. Of the 1338 patients, 413 (30.9 percent) had at least one such event. The incidence of severe adverse events was 13.8 percent in the placebo group, 15.2 percent in the 500-mg group, and 19.9 percent in the 1500-mg group (P = 0.038). Patients receiving 1500 mg of zidovudine were 2.1 times more likely to have a severe adverse event than those receiving placebo (P<0.001; 95 percent confidence interval, 1.6 to 2.8), whereas patients receiving 500 mg of zidovudine were 1.4 times more likely to experience an adverse event (P = 0.037; 95 percent confidence interval, 1.01 to 1.9). This dose-response relation was evident when the adverse events were broken down according to type (laboratory finding or symptom), with odds ratios of 2.2 for adverse laboratory findings in the 1500-mg group as compared with the placebo group (P<0.001; 95 percent confidence interval, 1.5 to 3.3) and 1.8 for the 500-mg group as compared with the placebo group (P = 0.003; 95 percent confidence interval, 1.2 to 2.7). Hematologic toxicity in the 1500-mg group and metabolic toxicity in the 500-mg group accounted for most of the differences in laboratory findings. Similarly, patients in the 1500-mg group were 1.6 times more likely to have symptoms than those in the placebo group (P = 0.015; 95 percent confidence interval, 1.08 to 2.36), whereas patients receiving 500 mg were no more likely to have symptoms than those receiving placebo (P = 0.54; odds ratio, 1.1; 95 percent confidence interval, 0.77 to 1.58). Gastrointestinal symptoms, particularly in the 1500-mg group, accounted for most of the differences in symptoms.

Figure 2. Figure 2. Partitioned Kaplan-Meier Survival Plots for the Placebo Group (Panel A), the 500-mg Group (Panel B), and the 1500-mg Group (Panel C).

Three end points are shown: event-free survival (EFS), progression-free survival (PFS), and overall survival (OS). These curves divide the 18 months of follow-up into three periods, as indicated by the areas between the plots: the time without symptoms or toxicity (unshaded area), the time after an adverse event (lightly shaded area), and the time after the progression of disease (darkly shaded area).

A Q-TWiST analysis was performed to compare treatments in terms of the quality-of-life benefits related to a delay in disease progression and the costs due to adverse events. Only symptoms were included as adverse events in this analysis. Figure 2 shows the partitioning of the overall survival time for the placebo group, the 500-mg group, and the 1500-mg group.

Table 2. Table 2. Survival Means for the Components of Q-TWiST, According to Treatment Group.

During the first 18 months in the study, the average period from the first occurrence of an adverse event to the progression of disease was 1.6 months in the placebo group, 2.1 months in the 500-mg group, and 2.7 months in the 1500-mg group (Table 2). The average period after disease progression was 0.7, 0.3, and 0.4 months, respectively. Placebo recipients had an average of 15.7 months without symptoms of disease or toxicity, as compared with 15.6 months for the 500-mg group and 14.8 months for the 1500-mg group. When we incorporated severe or life-threatening laboratory findings into the Q-TWiST calculations, we found that placebo recipients actually had an advantage over both zidovudine groups in terms of the time without symptoms of disease or toxicity (14.6, 13.7, and 12.1 months for the placebo, 500-mg, and 1500-mg groups, respectively).

Figure 3. Figure 3. Threshold-Utility Analysis Comparing a 500-mg Dose of Zidovudine with Placebo.

The vertical axis shows the value of the time after an adverse event (u_AE), and the horizontal axis the value of the time after disease progression (u_Prog). The values for both range from 0 to 1, with a value of 1 indicating that the time is worth the same as the time without symptoms or toxicity, and a value of 0 indicating that the time is worth nothing. The solid line denotes the threshold (based on the values of the time after an adverse event and the time after disease progression) at which zidovudine and placebo result in equal values for the quality-of-life-adjusted time without symptoms or toxicity (Q-TWiST). The dashed line shows the 95 percent confidence band for the threshold (the confidence band for placebo is out of the possible range of utility values and hence is not shown). The lines divide the plot into three areas showing the utility values for which Q-TWiST was significantly longer in the zidovudine group than in the placebo group (unshaded area), the values for which Q-TWiST was longer but not significantly so in the zidovudine group (lightly shaded area), and the values for which Q-TWiST was longer but not significantly so in the placebo group (darkly shaded area). There were no utility values for which placebo provided a significantly longer Q-TWiST than zidovudine.

Since 500 mg is the currently recommended daily dose of zidovudine, the remaining discussion focuses on the comparison between the group of patients receiving this dose of the drug and the placebo recipients. Figure 3 shows the results of the threshold-utility analysis. The value of the time after an adverse event and before the progression of disease is represented by the utility coefficient u_AE, and the value of the time after disease progression is indicated by u_Prog, both being relative to the time without symptoms of disease or toxicity. The figure shows the range of possible values for each utility coefficient and indicates which pairs of utility values favor zidovudine and which favor placebo. Subjective judgments provide the weights for the components of Q-TWiST, influencing treatment comparisons.

A traditional efficacy analysis uses a delay in the progression of disease as the main end point and assigns a value of 1 to u_AE, which is equivalent to time without symptoms or toxicity, and a value of 0 to u_Prog, which is equivalent to death. At these values, 500 mg of zidovudine is significantly better than placebo at delaying disease progression.

In our analysis, we instead assumed that the value of the time after an adverse event is not equivalent to the value of time without symptoms of disease or toxicity but rather is associated with a reduced quality of life. The actual value of that time depends on the individual patient and the specific adverse event. The utility value represents the amount of time without symptoms of disease or toxicity that would be equivalent to a year in the specified state of health (after an adverse event or disease progression). For example, with the values from Table 2, the comparison between 500 mg of zidovudine and placebo is computed as follows: -0.08 + (u_AE)(0.6) - (u_Prog)(0.5). For a patient who would give up 1 month of life from a year after an adverse event (u_AE = 0.92) or 4 months from a year after disease progression (u_Prog = 0.68) to gain time without symptoms of disease or toxicity, 500 mg of zidovudine provides a gain of 0.132 quality-of-life-adjusted months (4 days) over placebo during a period of 18 months, which is a significant difference. For any value of u_AE equal to or greater than the value of u_Prog, 500 mg of zidovudine provides more months of quality-of-life-adjusted time without symptoms of disease or toxicity than does placebo.

As for the effect of base-line CD4+ cell counts on our results, among the patients with a low CD4+ count, therapy was associated with a delay in disease progression of 1.3 months, but this difference between the therapy and placebo groups was not significant, perhaps because of the small sample (n = 162). However, there were no utility values in the threshold-utility plot that showed a significant advantage for zidovudine. In the group of patients with a high CD4+ count (n = 1173), the results were essentially identical to those for the sample as a whole. (Base-line CD4+ counts were not available for three patients.) A proportional-hazards regression model was used to predict threshold-utility values corresponding to CD4+ cell counts of 200, 300, and 400 per cubic millimeter. This analysis resulted in practically identical plots at each level, showing that the base-line CD4+ cell count had a minimal effect on the quality-of-life-adjusted time without symptoms or toxicity.

In the United States, early initiation of treatment with zidovudine at a dose of 500 mg per day is currently recommended for asymptomatic HIV-infected patients with CD4+ cell counts below 500 per cubic millimeter. This recommendation is based on evidence that zidovudine delays the progression of disease and that the toxicity associated with the drug is minimal and easily tolerated^2,7. However, the magnitude of the difference in disease progression originally reported for asymptomatic patients receiving placebo or 500 mg of zidovudine was relatively small, albeit significant (38 of 428 placebo recipients had disease progression, as compared with 17 of 453 zidovudine recipients, for time-adjusted rates of 7.6 and 3.6 end points per 100 person-years of observation, respectively)². That study assessed neither the quality of life nor the costs associated with zidovudine therapy. Recent reports suggesting that the initiation of zidovudine treatment when patients are asymptomatic provides no survival benefit have called into question the value of early intervention^4,16.

We believe that integrating outcomes related to the quality of life with traditional clinical end points may clarify this controversy. According to our study, asymptomatic patients receiving 500 mg of zidovudine who considered the progression of disease less desirable than a severe adverse event had a better quality of life than patients receiving placebo. However, even for the patient who valued the time after a severe adverse event four times more than the time after the progression of disease, the quality-of-life-adjusted time gained with 500 mg of zidovudine was less than 1 week during a period of 18 months.

The current analysis evaluates treatment effects during 18 months of follow-up and cannot rule out the possibility that greater benefits for zidovudine in terms of the quality of life will emerge with a longer follow-up. However, zidovudine alone is likely to be used only for a period of 12 to 18 months, given evidence that its effects are transient,¹⁷ that survival may not be improved with early treatment,⁴ and that a switch to didanosine yields better results than long-term zidovudine alone⁶. Therefore, a consideration of the quality of life during the first 18 months of therapy is important in determining when to initiate prolonged treatment with a sequence of drugs in asymptomatic patients.

In our study, the entire time after the first occurrence of a severe adverse event was considered to be part of the adverse experience. Although not all this time was actually associated with a poor quality of life, we believed that the occurrence of one of these events had the potential for a considerable effect on the patient. Unfortunately, since reliable information was not available about the duration of adverse events, we could not precisely determine the degree of reduction in the quality of life over time. Furthermore, in clinical practice, dose modifications and discontinuation of therapy may minimize the continuing negative effect of adverse events. We assumed, however, that the quality of life was generally lower (although fluctuating) after the initial adverse event and that the average degree of reduction in the quality of life could be represented by the utility coefficient. Counting all the remaining time after the occurrence of the initial adverse event as part of the adverse experience may have somewhat biased the analysis against zidovudine therapy.

Our main analysis focused on severe or life-threatening adverse events because of their substantially negative effect on the quality of life. Since the toxic effects of zidovudine tend to occur before the disease progresses, patients must be willing to tolerate adverse events in the near future for the sake of delaying the progression of disease, which is a benefit in the more distant future. It is possible, however, that the limited quality-of-life benefit of early initiation of zidovudine therapy would be outweighed by the occurrence of grade 3 or grade 4 laboratory findings, associated mild or moderate symptoms, anxiety over test results, or more intense medical management. Furthermore, grade 1 or 2 symptoms, which were not included in the model but were experienced by nearly all the patients, may be sufficiently disturbing to outweigh the gain associated with a delay in the progression of disease, particularly for asymptomatic patients. In the absence of a demonstrated survival benefit, a reasonable option for HIV-infected asymptomatic patients may be to delay the initiation of antiretroviral therapy until there is clinical or laboratory evidence of disease progression^18,19.

These findings suggest that clinicians should attempt to incorporate patients' preferences into treatment recommendations, instead of making such decisions solely on the basis of the CD4+ cell count and published guidelines. Future research may establish utility values for u_AE and u_Prog that are based on patients' preferences, so that information on the quality of life can be integrated into traditional efficacy trials. In addition, a direct assessment of the quality of life should be included in future clinical trials of treatment for HIV infection.